Dify Workflow and Chatflow: Which Agentic App Type to Build

Who this is for, and what you will be able to do

This guide is for people who want to build an AI application in Dify that runs a defined sequence of steps, and who are unsure which app type to start with. After reading it, you will know which type fits a task, what the start and end nodes are, and how nodes are connected.

Why use an agentic workflow

AI models on their own can be unpredictable. They may hallucinate, skip steps or produce inconsistent output. An agentic workflow puts AI capabilities inside a structured, repeatable process: it orchestrates models, tools and logic step by step, with explicit conditions, checkpoints and fallback paths. The AI still does the heavy lifting, but inside boundaries you define^[1].

Workflow or Chatflow

Dify provides two app types for building these flows: Workflow and Chatflow^[1]. Both are built on the same visual canvas and node system^[1].

• Workflow runs once from start to finish. It receives input, processes it through the flow and returns a result. Tasks such as automated report generation, data processing pipelines and batch processing fit this shape^[1].
• Chatflow adds a conversation layer. Users interact through a chat interface, and each message triggers the flow you designed before a reply is generated. Interactive assistants and guided question answering fit this shape. Chatflow also supports optional features such as content moderation and text-to-speech^[1].

A simple rule follows from this. If the input arrives once and you need one result back, choose Workflow. If the input is an ongoing conversation and every turn needs structured processing, choose Chatflow^[1].

How the start and end nodes differ

| | Workflow | Chatflow | |:---|:---|:---| | Start node | User Input or Trigger | User Input | | End node | Output (optional) | Direct Reply (required) |

The table reflects the node rules for each type^[1]. A Workflow can be started by a trigger, which runs it on a schedule, through a webhook or when an integration event occurs^[1]. A Chatflow always starts from a user message, so it has no triggers^[1].

Building a flow by connecting nodes

Each node handles one specific step, such as calling a model, retrieving knowledge, running code or branching on a condition^[1]. Building a flow is mostly a matter of dragging nodes onto the canvas, connecting them and configuring each one. Code is needed only when no existing node covers the logic you need^[1].

Build order

1. Choose the app type: Workflow for a one-time run that returns a result, or Chatflow for a conversation in which each message triggers the flow^[1].
2. Start with the start node: User Input for either type, or a Trigger for a Workflow^[1].
3. Add a node for each step, such as calling a model, retrieving knowledge, running code or branching on a condition, and connect them on the canvas^[1].
4. Finish with the end node: an optional Output node for a Workflow, or a required Direct Reply node for a Chatflow^[1].

Each node is configured separately. Code is needed only when no existing node covers the logic you need^[1].

Where to go next

How each app type orchestrates its logic is covered on a separate orchestration logic page^[1]. The full set of Dify documentation is the place to go for details beyond this overview^[1].

Adapted from the Dify documentation page *Workflow and Chatflow*^[1], released under the CC BY 4.0 license.